Regional Temperature Trends and Future Warming Risks in Pakistan Using Bias‐Corrected CMIP6 Ensembles and Machine Learning Techniques

ABSTRACT

Pakistan faces climate‐related hazards such as floods, droughts, and heatwaves, driven by diverse climate and rising temperatures. Reliable predictions are crucial for water management, agriculture, and disaster response, but the country's complex terrain requires region‐specific assessments. This study improves regional temperature projections by integrating ensembling techniques, statistical bias correction, and machine learning algorithms. Fuzzy C‐Means clustering divided Pakistan into six climatically similar regions. CMIP6 model outputs were evaluated and bias‐corrected to reduce errors. Temperature trends were estimated using both Sen's slope estimator and Random Forest regression (RFR). Sen's slope identified steady warming trends, while RFR captured nonlinear changes more effectively. Future projections under the SSP2‐4.5 and SSP5‐8.5 scenarios for three periods (2015–2044, 2045–2074, and 2075–2100) show significant warming, especially in central and southern Pakistan. Spatial analysis shows that warming increases relative to the baseline, notably under SSP5‐8.5, with larger increases in lowland and southern areas. The SSP2‐4.5 scenario suggests partial stabilisation toward the end of the century, while the SSP5‐8.5 scenario predicts accelerated warming that threatens food security, water resources, and public health. Bias correction reduces systematic deviations between modelled and observed data, but uncertainties persist under non‐stationary climate conditions. The framework improves regional analysis, yet uncertainties arising from bias‐correction assumptions and model limitations should be considered when interpreting results. By integrating these techniques, the study fills a gap in Pakistan's climate research, offering localised insights into future warming and valuable guidance for climate adaptation and policymaking.